About these questions
These five questions are original, one for each of the five AIF-C01 domains. We wrote them in the style of the AWS exam guide objectives. They are not real exam questions, and the real exam will word things differently. You can read more about how we build questions on our how we write our questions page.
How to use these
Try each question without looking at your notes, and pick an answer before you open the explanation. Then read the explanation even when you were right, because the reason the closest wrong option fails is where the learning is. Note any domain where you slipped and revisit it on the domains explained page.
Question 1 (Fundamentals of AI and ML)
A logistics company re-scores the delay risk of 25 million shipment records every weekend. All records are already stored in Amazon S3 by Friday night, nobody waits for individual results, and the report is needed on Monday morning. Which inference approach fits best?
- A. A real-time endpoint that stays provisioned all week
- B. Serverless inference called once for each record
- C. A batch inference job that reads the records from Amazon S3 and writes the scores back to Amazon S3
- D. Asynchronous inference that queues each request on a hosted endpoint
Show the answer and explanation
Correct answer: C. The complete input set exists before the run starts and results are only needed by morning, so a batch job is the natural match. Compute is used for the job and then released. Asynchronous inference, the closest wrong option, is meant for large individual requests that arrive live and can wait for a result, not for one bulk job over stored data. A real-time endpoint would pay for idle capacity for most of the week, and serverless inference suits interactive traffic that comes in bursts.
Question 2 (Fundamentals of GenAI)
An internal assistant built on a foundation model is now live for all employees. Which two activities belong to the deployment and feedback stage of the foundation model lifecycle? Choose two.
- A. Tracking answer quality, latency, and cost on production dashboards
- B. Selecting and cleaning the text collection used for pre-training
- C. Deciding on the architecture and the number of parameters
- D. Reviewing employee ratings on answers to plan improvements
- E. Running the initial pre-training job
Show the answer and explanation
Correct answers: A and D. Once an application is live, the work shifts to watching how it behaves and learning from users. Monitoring quality, latency, and cost, and collecting ratings that feed the next iteration, are both part of that stage. Selecting and cleaning the pre-training data belongs to data selection, and choosing the architecture belongs to model design, so B and C come much earlier. Running the first pre-training job also happens long before any employee sees the assistant, which makes E wrong as well.
Question 3 (Applications of Foundation Models)
An insurance chatbot starts every request to Amazon Bedrock with the same 6,000-word policy handbook, followed by a short customer question. The team wants lower cost and faster first responses without shortening the handbook. Which feature helps most?
- A. Provisioned Throughput
- B. A model evaluation job
- C. Batch inference
- D. Prompt caching
Show the answer and explanation
Correct answer: D. When many requests share the same long opening block, that block would otherwise be processed again on every call. Prompt caching keeps the processed prefix so later requests reuse it, which lowers input cost and speeds up the first part of the response. Provisioned Throughput, the closest wrong option, reserves model capacity for steady traffic, but each request would still reprocess the whole handbook. Batch inference is for offline workloads, and a model evaluation job measures quality instead of improving how requests are served.
Question 4 (Guidelines for Responsible AI)
A travel company’s assistant must never give visa or legal advice, even when customers ask in an indirect or hypothetical way, such as “suppose someone stayed past their permit, what would happen?” Which Amazon Bedrock Guardrails policy addresses this most directly?
- A. A word filter with a list of blocked terms
- B. A sensitive information filter that masks personal data
- C. A contextual grounding check against source passages
- D. A denied topics policy that describes visa and legal advice as out of scope
Show the answer and explanation
Correct answer: D. A denied topic is defined by a description of the subject in plain language, so the guardrail can catch indirect and hypothetical phrasings and not only exact words. A word filter, the closest wrong option, only matches the terms on its list, and a customer can avoid them easily. The sensitive information filter deals with personal data such as names and numbers, and a contextual grounding check measures whether an answer is supported by its sources. Neither of those decides which subjects the assistant may discuss.
Question 5 (Security, Compliance, and Governance for AI Solutions)
A support team pastes past tickets that contain customer names and phone numbers into prompts as examples. The security officer worries that these values could be repeated in answers or end up in logs. Which action addresses this most directly?
- A. Add more example tickets to the prompt
- B. Move the prompt logs to Amazon S3 Glacier
- C. Redact or tokenize the sensitive fields before the prompt is built, and mask them in outputs and logs
- D. Lower the model’s temperature setting
Show the answer and explanation
Correct answer: C. Data leakage is best prevented by keeping the sensitive values out of the prompt in the first place and by masking them wherever responses and logs are stored. A guardrail with a personal information policy can add a second layer. Lowering the temperature, the closest wrong option, only makes the wording less varied, and it does nothing to hide values that are already in the prompt. Archiving logs to Glacier keeps the same data in storage, and adding more tickets would expose even more customer details.
More guides for AIF-C01
- AIF-C01 Exam Format, Scoring and Retakes
- AIF-C01 Prerequisites and Experience
- AIF-C01 Renewal and Retakes Explained
- Is AIF-C01 Worth It? An Honest Look
- AIF-C01 Cheat Sheet: Key Facts to Review
- AIF-C01 vs MLA-C02: Which First?
- AWS AIF-C01 vs AI-901: Which First?
Next step
Check where you stand with the study plan, then look at the is it hard page for a readiness check. For more questions, our AIF-C01 practice tests are five original practice sets built around the five domains.